Exploring the Drivers of Content Entrepreneurs’ Compliance with Generative AI Policies: A Mixed-Methods Approach

Abstract

Social media-based content entrepreneurship is evolving rapidly and emerging as a significant and growing form of employment. Generative AI (GenAI) offers content entrepreneurs a powerful tool for content creation; however, the technology can be abused to produce deepfakes, rumors, plagiarism, and other injurious content. This triggers value co-destruction across the creator economy and society, making it particularly crucial to enhance content entrepreneurs’ compliance with GenAI policies. Aiming to develop an effective governance framework, this study adopts a mixed-methods approach, beginning with exploratory interviews to uncover factors affecting GenAI policy compliance intention. Subsequently, it employs confirmatory quantitative research with a survey to validate the proposed research model. The results indicate that both the deterrence triad (i.e., perceived sanction certainty, severity, and celerity) and perceived social norm strengthen GenAI policy compliance intention. Meanwhile, perceived social norm weakens the impact of perceived sanction certainty on policy compliance intention. Furthermore, peer communication enhances policy compliance intention by increasing perceptions of sanction certainty and celerity as well as social norm. These findings contribute to the sustainable development of content entrepreneurship and effective GenAI governance, fostering a symbiotic creator economy.

1. Introduction

Content entrepreneurship, a subset of digital entrepreneurship, denotes individuals creating and distributing valuable content on digital platforms to build an audience and generate revenue [1,2]. The growing demand for content consumption has attracted an increasing number of people to engage in this field, effectively driving new forms of employment. On a global level, approximately 200 million people worldwide are considered content creators, with a significant majority (66%) pursuing it as a side hustle, indicating a vast and expanding market [3]. This trend is particularly evident in China, where the number of online content entrepreneurs had surged to 15.08 million by the end of 2023 [4]. Therefore, the Chinese context, with its massive and rapidly growing creator economy, presents a critical case for examining the dynamics and challenges of content entrepreneurship.

Within this vast and dynamic market, generative AI (GenAI) has emerged as a pivotal force shaping its evolution. With its advanced data processing and self-learning capabilities, GenAI is now revolutionizing content entrepreneurship at an unprecedented pace across fields like intelligent writing, image generation, speech synthesis, and virtual character creation [5,6]. However, like any emerging technology, the thriving development of GenAI is accompanied by a series of abuse risks that cannot be overlooked. For instance, some content entrepreneurs abuse GenAI to mass-produce vulgar, deepfake, and/or deliberately misleading controversial content, which spreads rapidly across social media platforms, misleading public opinion and disrupting social order [7,8]. Therefore, the double-edged sword nature of GenAI not only optimizes content creation processes but also poses new requirements for the roles and responsibilities of content entrepreneurs [9]. However, research on this issue remains strikingly limited. In light of these considerations, developing approaches to steer content entrepreneurs toward responsible GenAI adoption and usage is critical—not only for empowering content entrepreneurship but also for providing platforms and policymakers with actionable evidence to formulate effective regulations. Therefore, this study addresses the following core research question: How can content entrepreneurs’ compliance with GenAI policies be enhanced to achieve effective governance of GenAI-driven content entrepreneurship?

Deterrence theory, originating in the works of Jeremy Bentham (1748–1832) and Cesare Beccaria (1738–1794), provides a robust framework for addressing the current research question, having been empirically validated within the Management Information Systems (MIS) domain. Studies have successfully applied it to identify factors that enhance employee compliance with information security policies [10,11] and curb technology misuse/abuse [12,13]. However, a thorough comparison of extant research uncovers a critical issue: a lack of consensus on the core components of deterrence, coupled with persistent inconsistencies in empirical findings (for full details, see Table 1). This discrepancy stems primarily from divergent research foci across studies, which have variably operationalized deterrence theory’s core constructs, ranging from sanction severity alone [11], to dyads of sanction certainty and severity [14,15], and finally triadic combinations incorporating sanction certainty, severity, and celerity [16]. More notably, these constructs yield different or even contradictory empirical results [15,17]. Although researchers have offered reasonable explanations for these issues, the inconsistencies inevitably raise concerns for subsequent research, motivating scholars to more deliberately delineate deterrence theory’s constructs and identify potential interacting factors.

To address these challenges, this study adopts a mixed-methods approach to delineate contextual drivers of GenAI policy compliance intention, following Venkatesh, Brown, and Bala [18] and Venkatesh, Brown, and Sullivan [19]. First, exploratory qualitative research involving interviews with content entrepreneurs utilizing GenAI for content creation identifies potential factors influencing GenAI policy compliance intention, grounded in the deterrence theory. Second, building on these exploratory findings, a subsequent survey-based quantitative study validates the conceptual model derived from the qualitative phase. Given the methodological rigor of mixed-methods research, this study addresses the conceptual and empirical inconsistencies in deterrence theory and advances academic understanding by theoretically and empirically elucidating the GenAI policy compliance mechanisms within this emerging context. Importantly for practice, it highlights how peer communication strengthens deterrence effectiveness, shapes social norms, and ultimately enhances policy compliance intention, thereby offering actionable insights for policymakers and platforms to design closed-loop governance frameworks.

The subsequent sections proceed as follows: Section 2 reviews literature on content entrepreneurship, GenAI, and deterrence theory to establish theoretical groundwork. Section 3 details exploratory qualitative research through semi-structured interviews with content entrepreneurs who use GenAI for content creation. Section 4 presents confirmatory quantitative analysis using survey data and partial least squares structural equation modeling (PLS-SEM) to validate the conceptual framework. Section 5 consolidates findings, discusses theoretical and practical implications, acknowledges limitations, and proposes future research directions.

2. Theoretical Background

2.1. Content Entrepreneurship

Content entrepreneurship refers to an entrepreneurial practice where content creators attract and retain specific audience groups by creating and disseminating valuable content (e.g., text, images, audio, video, etc.) on social media platforms, subsequently realizing commercial value through advertising, paid subscriptions, brand endorsements, and other means [20,21,22]. With the widespread adoption of social media platforms and the democratization of content creation tools, content entrepreneurship has emerged as a significant entrepreneurial form in the digital era, fueling the rise of the creator economy [2].

Existing research primarily employs conceptual approaches to explore how content entrepreneurs leverage their expertise to create and disseminate diverse content through social media platforms [23,24]. Nevertheless, amid its rapid evolution, content entrepreneurship continues to confront substantial challenges in practice. First, it necessitates that entrepreneurs possess not only talent and skills but also the capacity to effectively manage relationships with platforms, partner brands, and consumers [24]. Second, producing high-quality content to attract fans is cornerstone of successful content entrepreneurship, but it requires a significant investment of time and effort [25]. Finally, the revenue source is relatively unstable, which leads some content entrepreneurs to abandon their authenticity for monetary gains, ultimately falling into the “creator dilemma” [22]. To address these challenges, an increasing number of content entrepreneurs are turning to GenAI for the strategic enhancement of their content creation [26]. However, as GenAI is a double-edged sword, some content entrepreneurs may abuse it to pursue unethical profits, causing value co-destruction within the ecosystem [9,27]. Therefore, their adherence to GenAI policies is imperative to mitigate these associated risks [8]. Given that the integration of GenAI into content entrepreneurship is still in its nascent stage, significant governance gaps exist that merit further examination.

2.2. Generative AI

GenAI is an artificial intelligence technology that utilizes deep learning algorithms to generate new content by learning from vast amounts of training data [28]. It can automatically create various forms of content, such as text, images, audio, and video, based on user instructions, significantly enhancing content creation efficiency [6]. Compared to traditional content creation methods, GenAI offers notable advantages. It can rapidly generate large volumes of diverse content that meets market demands, serving as a strategic tool for content entrepreneurs to seek inspiration and produce premium content [5].

Despite its significant creative potential, GenAI adoption faces multifaceted challenges and risks arising from technical constraints and ethical uncertainties [29]. A primary concern is AI hallucination, where distortions erode information integrity and mislead users [30]. Further, individual use of GenAI may lead to intellectual property infringement and/or socio-ethical violations [8]. More critically, GenAI can be maliciously manipulated to generate false information and deepfake content, undermining social trust and information security [28]. These risks are particularly pronounced in the creator economy, posing a dual threat to the sustainability of the content ecosystem and social trust [9].

To mitigate these challenges, stakeholders—including GenAI developers, digital platforms, and governments—are actively formulating a multi-layered governance framework. For instance, OpenAI employs content filters in ChatGPT to manage queries and ensure the generation of accurate content [29]. Digital content platforms are refining algorithms and policies to detect and mitigate GenAI-generated disinformation [31]. Meanwhile, global leaders in AI—such as China, the United States, and the European Union—are developing comprehensive regulatory frameworks to address GenAI risks and align its deployment with ethical standards [32]. Current initiatives by various stakeholders primarily emphasize technological solutions and top-level regulations, yet lack in-depth exploration of user behavior at the application level. Therefore, investigating content entrepreneurs’ compliance with GenAI policies emerges as a critical research priority to ensure the effectiveness of these governance initiatives.

2.3. Deterrence Theory

Deterrence theory, originating in the 18th-century works of Jeremy Bentham (1748–1832) and Cesare Beccaria (1738–1794), posits that individuals make rational cost–benefit decisions on whether to engage in criminal or rule-violating acts [33]. It further holds that such behavioral motivation decreases when individuals perceive high detection probability (sanction certainty), severe penalties (sanction severity), and swiftness of punishment (sanction celerity) [33,34]. As humanity entered the information society, the proliferation of information technologies (ITs) has boosted human productivity while also giving rise to new forms of deviant behavior, such as computer abuse, which has resulted in substantial organizational losses. In response, MIS scholars have introduced deterrence theory to address this issue. Representatively, Straub [35] applied general deterrence theory to reveal that sanction certainty and sanction severity significantly reduce employees’ computer abuse. This foundational work catalyzes subsequent research that advances the deterrence theory through refinement of core constructs and empirical validation via diverse methodologies. Table 1 lists representative studies applying deterrence theory to technology usage policy compliance.

Research in this vein reveals four dominant patterns regarding employee IS policy compliance. First, findings on core deterrence components (i.e., sanction certainty, severity, celerity) and their effects on policy compliance show marked inconsistency. For example, while Straub [35] and Chen et al. [11] confirm that sanction certainty and severity positively influence compliance; Li, Zhang, and Sarathy [10] find severity ineffective, and Herath and Rao [17] even show it reduces compliance intention. As for sanction celerity, Johnston, Warkentin, and Siponen [36] confirm it fails to enhance compliance intention. This variability, as Trang and Brendel’s [37] meta-analysis confirms, is highly context-dependent—the most prominent characteristic of deterrence theory in its application and evolution, and one that merits special attention. Second, scholars refine the deterrence theoretical construct by distinguishing between informal and formal sanctions, finding that informal sanctions effectively enhance compliance intention whereas formal sanctions prove ineffective [36]. Third, researchers integrate deterrence theory with other theoretical perspectives to develop comprehensive models to better explain policy compliance; yet results demonstrate that deterrence mechanisms mostly lose effectiveness in such frameworks [13,15,38,39,40,41]. Fourth, studies are increasingly incorporating critical antecedents—such as security countermeasures [12], awareness of being monitored [16], and emotional states [14]—into their frameworks, with the aim of strengthening the causal pathways underlying deterrence theory.

Table 1. Deterrence theory applications in MIS.

Study	Model Summary				Research Method	Major Finding(s)
	Independent Variable(s)	Mediator(s)	Moderator(s)	Dependent Variable(s)
[35]	Deterrent certainty, Deterrent severity, Preventive security software, Motivational factors, Environmental factors	-	-	Computer abuse	Survey	General deterrence theory in criminology serves as the conceptual foundation for understanding and mitigating computer abuse in organizational settings. Deterrent certainty and severity significantly mitigate employee computer abuse. Preventive security software leads to reduced computer abuse.
[12]	Security policies, SETA program, Computer monitoring	Perceived sanction certainty, Perceived sanction severity	-	IS misuse intention	Hypothetical scenario research design	Security policies, SETA program, and computer monitoring can significantly affect perceived sanction severity that decreases IS misuse intention. Security policies, SETA program, and computer monitoring can significantly affect perceived sanction certainty; however, perceived sanction certainty exerts no significant effect on IS misuse intention.
[17]	Severity of penalty, Certainty of detection, Normative beliefs, Peer behavior	-	-	IS security policy compliance intention	Survey	Certainty of detection enhances policy compliance intention, whereas severity of penalty decreases policy compliance intention. Normative beliefs and peer behavior both strengthen policy compliance intention.
[10]	Detection probability, Sanction severity, Security risks, Perceived benefits, Personal norms (Moral beliefs)	-	Personal norms	Internet usage policy compliance intention	Survey	Detection probability increases policy compliance intention. Sanction severity cannot affect policy compliance intention. Benefits of Internet abuse decreases policy compliance intention. Personal norms increase policy compliance intention. Personal norms negatively moderate the effect of sanction severity on policy compliance intention.
[38]	Neutralization techniques, Formal/informal sanctions, Shame	-	-	Intention to violate IS security policy	Hypothetical scenario research design	Formal and informal sanctions cannot significantly influence intention to violate IS security policy. Neutralization techniques can significantly affect intention to violate IS security policy. Shame cannot significantly affect intention to violate IS security policy.
[36]	Formal/informal sanction certainty, Formal/informal sanction severity, Sanction celerity Perceived threat severity, Perceived threat susceptibility	Perceived self-efficacy, Perceived Response efficacy	-	IS security policy compliance intention	Hypothetical scenario research design	Informal sanction severity and informal sanction certainty can enhance policy compliance intention, whereas formal sanction severity and formal sanction certainty have no such effects. Sanction celerity cannot significantly affect policy compliance intention. Perceived self-efficacy and perceived response efficacy, respectively, partially mediate the effect of perceived threat severity on policy compliance intention.
[39]	Neutralization techniques, Formal/informal sanction certainty, Formal/informal sanction severity	Shame, Intention to use shadow IT	-	Actual usage of shadow IT	Survey	Formal/informal sanction severity as well as formal/informal sanction certainty are not significant predictors of intention to use shadow IT. Informal sanction certainty and severity are predictors of shame. “Metaphor of the ledger” neutralization technique predicts intention to use shadow IT.
[11]	Perceived sanction severity	Perceived self-efficacy, Perceived descriptive norm, Perceived response cost	Perceived self-efficacy, Perceived descriptive norm, Perceived response cost	IS security policy compliance intention	Survey	Perceived sanction severity increases IS security policy compliance intention. Perceived self-efficacy, perceived descriptive norm, and perceived response cost fail to interact with perceived sanction severity to influence policy compliance intention. Perceived self-efficacy and perceived descriptive norm mediate the effect of perceived sanction severity on policy compliance intention.
[37]	Formal/informal sanction certainty, Formal/informal sanction severity, Sanction celerity	-	Contextual moderators, Methodological moderators	Information security policy compliance behavior	Meta-analysis	The effectiveness of sanction factors in affecting policy compliance shows significant variability across studies, depending on the study context. The results from the scenario-based and behavior-specific measurements show no significant methodological differences.
[16]	Awareness of being monitored	Sanction severity, Sanction certainty, Sanction celerity	-	Computer usage policy compliance intention	Survey	Sanction severity and sanction certainty can mediate the effects of awareness of monitored on policy compliance intention, whereas sanction celerity fails to play such role.
[13]	Formal/informal sanction certainty, Formal/informal sanction severity, Shame, Moral beliefs, Neutralization techniques	-	Power distance, Uncertainty avoidance, Individualism vs. Collectivism	Intention to violate IS security policy	Hypothetical scenario research design	Formal/informal sanction certainty and severity cannot decrease employees’ intention to violate IS security policy. Shame and moral beliefs decrease employees’ intention to violate IS security policy. Neutralization techniques increase employees’ intention to violate IS security policy.
[14]	Fear, Anger	Formal/informal sanction certainty, Formal/informal sanction severity	-	Computer-related deviant behavioral intention	Hypothetical scenario research design	Experienced fear enhances perceived formal/informal sanctions which then decrease computer-related deviant behavioral intention. Experienced anger reduces perceived informal sanctions which then decrease computer-related deviant behavioral intention.
[15]	Perceived deterrent severity, Perceived deterrent certainty, Ethical leadership, Abusive supervision	-	-	IS security policy compliance intention	Hypothetical scenario research design	Perceived deterrent certainty strengthens policy compliance intention, whereas perceived deterrent severity fails to exhibit significant effect. Ethical leadership enhances policy compliance intention, whereas abusive supervision fails to exhibit such effect.
[40]	Organizational sanctions (sanction severity, sanction certainty, sanction celerity), Financial benefits, Self-control, Psychological contract violations	Organizational deterrence	Organizational deterrence, Self-control	Insider computer abuse	Survey	Organizational sanctions can promote organizational deterrence. Organizational deterrence cannot directly decrease insider computer abuse, but can attenuate the positive effect of perceived financial benefits on insider computer abuse. Self-control not only directly decreases insider computer abuse, but also attenuates the positive effects of perceived financial benefits and psychological contract violations on insider computer abuse.
[41]	Formal/informal sanction severity, Formal/informal sanction certainty, Shame, Moral beliefs, Neutralization techniques	-	Gender	University students’ intention to misuse ChatGPT	Hypothetical scenario research design	All neutralization techniques enhance university students’ ChatGPT misuse intention. Formal sanction severity decreases university students’ ChatGPT misuse intention. Moral beliefs decrease university students’ ChatGPT misuse intention. Gender demonstrates moderating effects.
This study	Peer communication	Perceived sanction certainty, severity, and celerity, Perceived social norm	Perceived social norm	Content entrepreneurs’ intention to comply with GenAI policies	Mixed-method approach	Perceived sanction certainty, severity, and celerity enhance GenAI policy compliance intention. Perceived social norm strengthens GenAI policy compliance intention. Perceived social norm weakens the effect of perceived sanction certainty on GenAI policy compliance intention. Peer communication enhances perceived sanction certainty and celerity, but not perceived sanction severity. Peer communication increases perceived social norm.

Table 1. Deterrence theory applications in MIS.

Study	Model Summary				Research Method	Major Finding(s)
	Independent Variable(s)	Mediator(s)	Moderator(s)	Dependent Variable(s)
[35]	Deterrent certainty, Deterrent severity, Preventive security software, Motivational factors, Environmental factors	-	-	Computer abuse	Survey	General deterrence theory in criminology serves as the conceptual foundation for understanding and mitigating computer abuse in organizational settings. Deterrent certainty and severity significantly mitigate employee computer abuse. Preventive security software leads to reduced computer abuse.
[12]	Security policies, SETA program, Computer monitoring	Perceived sanction certainty, Perceived sanction severity	-	IS misuse intention	Hypothetical scenario research design	Security policies, SETA program, and computer monitoring can significantly affect perceived sanction severity that decreases IS misuse intention. Security policies, SETA program, and computer monitoring can significantly affect perceived sanction certainty; however, perceived sanction certainty exerts no significant effect on IS misuse intention.
[17]	Severity of penalty, Certainty of detection, Normative beliefs, Peer behavior	-	-	IS security policy compliance intention	Survey	Certainty of detection enhances policy compliance intention, whereas severity of penalty decreases policy compliance intention. Normative beliefs and peer behavior both strengthen policy compliance intention.
[10]	Detection probability, Sanction severity, Security risks, Perceived benefits, Personal norms (Moral beliefs)	-	Personal norms	Internet usage policy compliance intention	Survey	Detection probability increases policy compliance intention. Sanction severity cannot affect policy compliance intention. Benefits of Internet abuse decreases policy compliance intention. Personal norms increase policy compliance intention. Personal norms negatively moderate the effect of sanction severity on policy compliance intention.
[38]	Neutralization techniques, Formal/informal sanctions, Shame	-	-	Intention to violate IS security policy	Hypothetical scenario research design	Formal and informal sanctions cannot significantly influence intention to violate IS security policy. Neutralization techniques can significantly affect intention to violate IS security policy. Shame cannot significantly affect intention to violate IS security policy.
[36]	Formal/informal sanction certainty, Formal/informal sanction severity, Sanction celerity Perceived threat severity, Perceived threat susceptibility	Perceived self-efficacy, Perceived Response efficacy	-	IS security policy compliance intention	Hypothetical scenario research design	Informal sanction severity and informal sanction certainty can enhance policy compliance intention, whereas formal sanction severity and formal sanction certainty have no such effects. Sanction celerity cannot significantly affect policy compliance intention. Perceived self-efficacy and perceived response efficacy, respectively, partially mediate the effect of perceived threat severity on policy compliance intention.
[39]	Neutralization techniques, Formal/informal sanction certainty, Formal/informal sanction severity	Shame, Intention to use shadow IT	-	Actual usage of shadow IT	Survey	Formal/informal sanction severity as well as formal/informal sanction certainty are not significant predictors of intention to use shadow IT. Informal sanction certainty and severity are predictors of shame. “Metaphor of the ledger” neutralization technique predicts intention to use shadow IT.
[11]	Perceived sanction severity	Perceived self-efficacy, Perceived descriptive norm, Perceived response cost	Perceived self-efficacy, Perceived descriptive norm, Perceived response cost	IS security policy compliance intention	Survey	Perceived sanction severity increases IS security policy compliance intention. Perceived self-efficacy, perceived descriptive norm, and perceived response cost fail to interact with perceived sanction severity to influence policy compliance intention. Perceived self-efficacy and perceived descriptive norm mediate the effect of perceived sanction severity on policy compliance intention.
[37]	Formal/informal sanction certainty, Formal/informal sanction severity, Sanction celerity	-	Contextual moderators, Methodological moderators	Information security policy compliance behavior	Meta-analysis	The effectiveness of sanction factors in affecting policy compliance shows significant variability across studies, depending on the study context. The results from the scenario-based and behavior-specific measurements show no significant methodological differences.
[16]	Awareness of being monitored	Sanction severity, Sanction certainty, Sanction celerity	-	Computer usage policy compliance intention	Survey	Sanction severity and sanction certainty can mediate the effects of awareness of monitored on policy compliance intention, whereas sanction celerity fails to play such role.
[13]	Formal/informal sanction certainty, Formal/informal sanction severity, Shame, Moral beliefs, Neutralization techniques	-	Power distance, Uncertainty avoidance, Individualism vs. Collectivism	Intention to violate IS security policy	Hypothetical scenario research design	Formal/informal sanction certainty and severity cannot decrease employees’ intention to violate IS security policy. Shame and moral beliefs decrease employees’ intention to violate IS security policy. Neutralization techniques increase employees’ intention to violate IS security policy.
[14]	Fear, Anger	Formal/informal sanction certainty, Formal/informal sanction severity	-	Computer-related deviant behavioral intention	Hypothetical scenario research design	Experienced fear enhances perceived formal/informal sanctions which then decrease computer-related deviant behavioral intention. Experienced anger reduces perceived informal sanctions which then decrease computer-related deviant behavioral intention.
[15]	Perceived deterrent severity, Perceived deterrent certainty, Ethical leadership, Abusive supervision	-	-	IS security policy compliance intention	Hypothetical scenario research design	Perceived deterrent certainty strengthens policy compliance intention, whereas perceived deterrent severity fails to exhibit significant effect. Ethical leadership enhances policy compliance intention, whereas abusive supervision fails to exhibit such effect.
[40]	Organizational sanctions (sanction severity, sanction certainty, sanction celerity), Financial benefits, Self-control, Psychological contract violations	Organizational deterrence	Organizational deterrence, Self-control	Insider computer abuse	Survey	Organizational sanctions can promote organizational deterrence. Organizational deterrence cannot directly decrease insider computer abuse, but can attenuate the positive effect of perceived financial benefits on insider computer abuse. Self-control not only directly decreases insider computer abuse, but also attenuates the positive effects of perceived financial benefits and psychological contract violations on insider computer abuse.
[41]	Formal/informal sanction severity, Formal/informal sanction certainty, Shame, Moral beliefs, Neutralization techniques	-	Gender	University students’ intention to misuse ChatGPT	Hypothetical scenario research design	All neutralization techniques enhance university students’ ChatGPT misuse intention. Formal sanction severity decreases university students’ ChatGPT misuse intention. Moral beliefs decrease university students’ ChatGPT misuse intention. Gender demonstrates moderating effects.
This study	Peer communication	Perceived sanction certainty, severity, and celerity, Perceived social norm	Perceived social norm	Content entrepreneurs’ intention to comply with GenAI policies	Mixed-method approach	Perceived sanction certainty, severity, and celerity enhance GenAI policy compliance intention. Perceived social norm strengthens GenAI policy compliance intention. Perceived social norm weakens the effect of perceived sanction certainty on GenAI policy compliance intention. Peer communication enhances perceived sanction certainty and celerity, but not perceived sanction severity. Peer communication increases perceived social norm.

Building on deterrence theory’s established role in domains like IT policy compliance, Goyal, Chauhan, and Motiwalla [41] have extended it into GenAI governance, demonstrating its utility in curbing misuse among university students. However, in the emerging field of content entrepreneurship, no research has yet applied deterrence theory to explain the impact of sanction elements on content entrepreneurs’ compliance with GenAI policies. Beyond this, two critical gaps persist in current research. First, identifying appropriate core deterrence components (i.e., sanction certainty, severity, and/or celerity) remains challenging due to inconsistent findings across prior studies. Second, the exploration of critical and novel content entrepreneurship-specific factors that amplify deterrent effects remains underexplored, which hinders the development of a closed-loop governance framework. To address these gaps, this study employs a mixed-methods approach centered on deterrence theory, combining qualitative interviews to identify context-specific deterrents and potential amplifying factors with a subsequent quantitative survey to validate these findings.

3. Exploratory Study

3.1. Research Method

Following the mixed-methods research guidelines developed by Venkatesh, Brown, and Bala [18] and Venkatesh, Brown, and Sullivan [19], this study first employs semi-structured interviews. These interviews, designed and conducted in accordance with deterrence theory, serve two primary objectives: (1) to identify the contextually relevant deterrence constructs pertinent to the current context and compare them with findings from prior research, and (2) to uncover novel and context-specific antecedents that influence these core deterrence elements. To better facilitate the interviews, four core questions have been designed: (1) In your experience, does GenAI abuse occur in content creation practices? (2) Have you engaged in non-compliant GenAI usage, or alternatively, adhered to responsible practices? (3) What factors influenced your decision to engage in the practices mentioned above? (4) What measures would you suggest to mitigate GenAI abuse and promote responsible GenAI use in this context?

To recruit interviewees, a recruitment poster was published on the Wenjuanxing platform using a convenience sampling approach. The poster explicitly required participants to be content entrepreneurs who actively utilize GenAI tools for content creation. Informed consent was obtained by clearly outlining on the poster the voluntary nature of participation, the purpose of the interviews, the interview process, and the information required from interviewees.

A total of 7 qualified participants were ultimately included in the exploratory interview study after rigorous identity verification. This sample size aligns with the minimum recommendation (i.e., more than 6 interviewees) put forth by Corbin and Strauss [42] and Gentles et al. [43] for qualitative studies aiming to uncover core concepts and categories. Meanwhile, theoretical saturation—defined as the point when new interviews no longer generate novel thematic insights—was systematically assessed in line with the method proposed by Corbin and Strauss [42]. As evidenced in

Table 2. Findings from the interviews.

# of Participant	Key Interview Quotes
#1 (Male, 22 years old, 2 years of content creation experience, primarily using deepseek)	“I have never abused GenAI. I’ve heard that the platform can detect certain types of GenAI-generated inappropriate content and impose penalties.”—(sanction certainty) “I’ve heard that most violations result in warnings initially. Then, if you fail to make necessary adjustments, you risk facing traffic restrictions or even being banned from the platform.”—(sanction severity) “My content is relatively niche… We often share experience with each other. Of course, now that GenAI exists, we also discuss how to utilize it more effectively.”—(peer communication) “Neither my friends nor I abuse GenAI, to my knowledge.”—(social norm)
	“I maintain my own professional red lines. I use GenAI but not abuse it, at least not now.”—(moral belief) “While abusing GenAI tools may generate short-term traffic spikes, consumers ultimately prefer high-quality content. Only the proper use of GenAI can ensure long-term benefits.”—(economic benefit) “I have viewed some inappropriate GenAI-generated content on the platform, such as parodying classic literary works like Journey to the West. I think the platform’s governance is inadequate and lacks systematic guidance and educational support for content creators.” —(lack of countermeasures)
#2 (Male, 56 years old, 6 months of content creation experience, primarily using deepseek and Doubao)	“My hobby is video editing. Once, I published a video edited with GenAI, and shortly after, the platform warned me of a violation. However, I didn’t know why.”—(sanction certainty, sanction celerity) “I complained about this experience in the community, only to find that others had similar experiences. Even some peers said certain accounts had been suspended for a while.”—(sanction certainty, sanction severity, peer communication) “In our community, we often discuss strategies for utilizing GenAI more effectively.”—(peer communication)
	“I think the platform lacks transparency in governance—it just flags your content without telling you exactly why. And I get the sense that macro-creators are treated more leniently, while micro-ones are held to a stricter standard.”—(lack of transparency in governance) “Competition for traffic is fierce now. So, I think occasionally using GenAI to imitate others’ content is acceptable as long as it’s not excessive.”—(neutralization techniques)
#3 (Female, 22 years old, 3 years of content creation experience, primarily using Doubao)	“There are certainly instances of GenAI abuse, but I believe the majority of people do not do this.”—(social norm) “I’ve seen news reports of individuals arrested for using GenAI to fabricate and spread disinformation.”—(sanction severity) “I primarily use GenAI to generate text scripts. As long as it’s used appropriately, the platform won’t intervene.”—(sanction certainty)
	“The use of GenAI that violates social ethics or disrupts public order is wrong.”—(moral belief) “Personally, I think platform governance remains inadequate. There’s a growing volume of GenAI-generated content that exhibits high homogeneity. Does this imply mutual plagiarism is acceptable?”—(lack of countermeasures)
#4 (Female, 39 years old, 3.5 years of content creation experience, primarily using KIMI and deepseek)	“There’s a large volume of GenAI-generated deceptive product recommendation content on the platform that attracts significant traffic. I also want to do this!”—(negative social norm) “But now, things have changed. The platform has initiated a stringent crackdown on GenAI-generated fraudulent content.”—(sanction certainty) “Within the community, I’ve seen content shared by peers stating that the platform introduces tiered penalties based on the severity of negative impact and the frequency of violations involving GenAI-generated deceptive content.”—(peer communication, sanction severity)
	“I also want to earn money, but I think it is morally wrong to abuse GenAI for attracting traffic; therefore, I have refrained from engaging in such practices.”—(economic benefits, moral belief) “The platform has issued several announcements stating that it will strengthen the governance of GenAI-generated content.”—(countermeasures)
#5 (Male, 30 years old, 5 years of content creation experience, primarily using Doubao)	“There is a great deal of GenAI-generated content on the platform. Some of it gains significant popularity, while some is of poor quality. In a word, an increasing number of people are using GenAI.”—(social norm) “Once I used GenAI to create content but was warned, possibly because the topic was sensitive.”—(sanction certainty) “Several times, the content was warned by the platform shortly after published. Possibly I used GenAI to imitate and repost others’ content.”—(sanction certainty, sanction celerity) “The platform is tightening its regulatory measures on GenAI-generated content, with penalties ranging from traffic throttling to account suspension. The requirements are getting more and more stringent. We’re all complaining about it!”—(sanction severity, peer communication) “I learn from others how to better utilize GenAI to create content and attract traffic. If others can, so can I.”—(peer communication, social norm)
	“Before the emergence of GenAI, the platform already had much inappropriate content—mainly for grabbing traffic. As GenAI appeared, the platform now has more fake, synthetic content. Platform supervision is not quite adequate.”—(lack of countermeasures)
#6 (Female, 48 years old, 10 months of content creation experience, primarily using deepseek and ERNIE)	“When I first started, I truly didn’t know how to use GenAI appropriately. I used it to imitate some content. Then, approximately two days later, the platform warned me to prove the content was original.”—(sanction certainty, sanction celerity) “The platform requires me to prove that the content is my original work; otherwise, my account will be throttled. That’s really troublesome! So, I have no choice but to play by its rules.”—(sanction severity) “I turned to my friends for help, and they told me a lot of practical advice.”—(peer communication)
	“I’m a newcomer with limited traffic, so sometimes I modify others’ popular content using GenAI. I know this isn’t right, but…”—(moral belief) “The platform’s traffic distribution mechanism is extremely unfair, compelling me to take such actions.”—(neutralization techniques)
#7 (Male, 26 years old, 5 months of content creation experience, primarily using deepseek and Doubao)	“I find many people are using GenAI to create popular content with high quality…. So, I think whether GenAI is abused or not is entirely up to the creators.”—(social norm) “My friend told me that abusing GenAI is risky: the platform will detect it, viewers may report it, and penalties will be imposed.”—(peer communication, sanction certainty) “I’ve learned from multiple sources that many accounts with a large number of followers have been banned by the platform, mainly due to the improper use of GenAI.”—(sanction severity)
	“As a newcomer, I use GenAI in a proper way to seek creative inspiration, which helps enhance content quality. I firmly believe that content quality is the key to gaining traffic.”—(economic benefit, moral belief) “I hope the platform can truly ensure the fairness of traffic distribution and encourage creators to post high-quality content.”—(countermeasures)

Note: The interview quotes in the white boxes are used to explore the research variables in this study. The quotes in the grey boxes are not incorporated into the research model, but they contribute to a more comprehensive understanding of the ways to improve content entrepreneurs’ compliance with GenAI policies. To avoid potential misunderstandings and bias, the names of four specific content platforms mentioned in the interviews are uniformly referred to as “the platform”.

, the later interviews consistently replicated the codes and themes identified in earlier ones, and no new concepts emerged. This observation thus confirms that theoretical saturation was achieved.

To address potential sampling biases while ensuring broad representation of perspectives, a demographically diverse cohort was formed to encompass the key characteristics of content entrepreneurs. As summarized in Table 2, participants varied in age (22–56 years), gender (4 males, 3 females), content creation experience (5 months to 5 years), types of GenAI tools used (e.g., DeepSeek, Doubao), and content distribution platforms (4 major social media platforms, anonymized to avoid misinterpretation). This diversity in participant backgrounds enabled the capture of a broad spectrum of viewpoints, thereby enriching qualitative insights into content entrepreneurs’ GenAI usage.

Interviews are conducted via Tencent Meeting. Each participant received ¥200 as compensation upon completion of their interviews. Two trained interview recorders (non-authors) document the interview content while the entire session is video-recorded. Each interview lasted approximately 40 min, during which the interviewees’ demographic information, as well as their responses to four preset core questions, were thoroughly recorded. To minimize potential transcription bias, this study implements two measures: First, each recorder cross-checks their manually transcribed interview records against AI software (i.e., iFLYrec)—generated transcripts. If discrepancies are identified, a co-author jointly reviews the video recording with that recorder to reach a final determination. Second, recorders exchange their completed transcripts for mutual comparison. When inconsistencies are detected during this exchange, all three parties (both recorders and one co-author) collectively re-examine the recording to verify content accuracy.

Subsequently, the interview data (the transcribed content) was annotated with codes using NVivo 15. In this process, this study referenced Schuetz et al. [44], which had predetermined goal systems theory as the theoretical framework that provided the researchers with general codes (i.e., concepts). Therefore, the researchers identified concepts specific to the research context and looked for common themes among the codes in this respect. Similarly to Schuetz et al. [44], this study predetermined deterrence theory as the theoretical framework. Thus, we coded the interview data around deterrence theory to identify concepts related to content entrepreneurs’ compliance with GenAI policies. Following coding principles form Corbin and Strauss [42], the coding was performed by two coders, each coder examined the interview transcripts line by line to identify codes (concepts) that could explain interviewees’ GenAI policy compliance. Given the relatively small sample size and the use of a theory-driven coding framework, a consensus-based coding strategy was adopted to ensure reliability, informed by Kim, Kankanhalli, and Lee [45]. Throughout this process, one author facilitated immediate discussion and resolution of discrepancies. This collaborative coding strategy not only guaranteed interrater reliability but also enhanced the precise identification of key concepts.

3.2. Interview Results

The interview findings are summarized in Table 2, which presents representative quotes aligned with key research concepts. First, the analysis reveals that sanction certainty, severity, and celerity are consistently emphasized by participants as fundamental to compliance behavior. This finding establishes the three elements as the core context-specific deterrence components for GenAI policy compliance, directly addressing the recognized challenge of identifying key deterrence dimensions across different technology-use contexts [37]. Second, peer communication emerges as a significant social process that shapes perceptions of policy enforcement, establishing it as a key potential amplifying factor. This variable has not been incorporated in prior MIS research applying deterrence theory [12,16], yet demonstrates particular significance in the context of content entrepreneurship. Third, while both social norm and moral belief are frequently mentioned as influencing compliance, this study specifically incorporates social norm due to its theoretical alignment with the study’s focus on social dynamics. According to established definitions, moral belief constitutes an individual’s internalized ethical stance that operates through personal conscience [10], whereas social norm represents perceived pressures from reference groups that function through external social influences [11]. Since peer communication functions as a distinctly social mechanism within content entrepreneur communities, social norm offers the appropriate theoretical basis for examining compliance transmission through group interaction processes. Consequently, the current framework excludes moral belief to preserve conceptual coherence with the social transmission mechanisms under investigation.

Furthermore, economic benefits, neutralization techniques, and countermeasures—though identified in interviews as relevant to GenAI compliance—are excluded from the research model to maintain theoretical coherence and parsimony. Regarding neutralization techniques and economic benefits, existing literature suggests that integrating neutralization theory and cost–benefit theory as competing theories alongside deterrence theory in a single framework may undermine the effectiveness of sanctions [13,39,40]. To maintain a clear theoretical focus on deterrence mechanisms, these constructs are therefore excluded. As for countermeasures, although they have been shown to influence perceived sanctions in previous studies [12,16], these variables exhibit substantial variations across different platforms, potentially introducing measurement biases in subsequent quantitative analysis. More importantly, compared to peer communication, countermeasures do not represent a novel construct in this context, and it remains theoretically challenging to position them alongside peer communication as concurrent influencing factors. Therefore, countermeasures are also excluded from the research model.

4. Quantitative Study

4.1. Theoretical Framework and Hypotheses

Based upon the qualitative study insights, this study develops the theoretical framework as shown in Figure 1. It first posits that when content entrepreneurs realize that their abuse of GenAI is almost inevitable to escape punishment (high certainty), that they will face severe penalties once discovered (high severity), and that the punishment will arrive promptly (high celerity), they are more likely to comply with the usage policies of GenAI.

Within digital content creation, GenAI policies mandate the generation of authentic, legally compliant, and ethically responsible content to mitigate risks like plagiarism, malicious deepfakes, and dissemination of misinformation [29]. Despite progressive refinement of these policies, GenAI abuse remains prevalent [46]. Thus, content platforms and regulatory bodies must effectively assume regulatory responsibilities, enhancing the deterrent effect against violations by increasing the certainty of sanction. Following Wang and Xu [15] and Burns et al. [40], this study defines perceived sanction certainty as content entrepreneurs’ subjective assessment of the likelihood that their GenAI policy violations will be detected and punished.

According to Sarkar [47], platforms equipped with integrated monitoring systems can implement end-to-end regulatory governance of GenAI-generated content. This governance encompasses pre-upload compliance screening, real-time authenticity validation, and post-publication traceability analysis. Fundamentally, this enhances sanction certainty by ensuring the unavoidable detection of violations. From a behavioral economics standpoint [17,48], content entrepreneurs are rationally compelled to calculate the expected costs of violating GenAI policies during content production when perceived sanction certainty is high. Meanwhile, as demonstrated by Wang and Xu [15] and Burns et al. [40], elevated sanction certainty fosters greater risk aversion among content entrepreneurs. Their recognition that violations are subject to high probabilities of detection and penalty elevates risk sensitivity, thereby motivating stricter compliance with GenAI policies. Synthesizing these arguments, this study advances the following hypothesis.

H1: 

Perceived sanction certainty for GenAI abuse positively affects content entrepreneurs’ intention to comply with GenAI policies.

Similarly, based on the research by Wang and Xu [15] and Burns et al. [40], this study defines perceived sanction severity as content entrepreneurs’ subjective assessment of the level of harshness associated with penalties for GenAI abuse. From an economic perspective, the severity of sanctions can deter potential violations or illegal behavior by increasing the cost of non-compliance [17,48]. Therefore, as rational economic agents, content entrepreneurs will inevitably conduct cost–benefit analyses weighing potential benefits against expected violation costs when deciding whether to violate GenAI policies. Through this calculus, the risk of severe sanctions substantially reduces the expected net utility of violations—potentially even rendering it negative. Consequently, this simultaneously diminishes the motivation for GenAI abuse and increases the intention to comply with policies.

Furthermore, from the perspective of reputation and long-term interests [49], reputation can be a crucial intangible asset for content entrepreneurs, suggesting they should place great importance on personal reputation management. Critically, reputation is hard-earned yet easily destroyed [49]. Heightened sanction severity carries a high risk of decimating content entrepreneurs’ painstakingly accumulated reputation, ultimately compelling them to comply with GenAI policies to preserve viable career sustainability. Based on the preceding arguments, the following hypothesis is derived.

H2: 

Perceived sanction severity for GenAI abuse positively affects content entrepreneurs’ intention to comply with GenAI policies.

In criminology, sanction celerity is the most overlooked element of deterrence theory. This is mainly because incorporating the swiftness of punishment into the criminal justice system is impractical [50]. Similarly, in the MIS field, many studies have not explored sanction celerity as a core element of deterrence theory [12,13]. Some studies have also shown that there is no significant relationship between sanction celerity and compliance with IT usage policies [16,36]. However, this study argues that in the context of the creator economy, with the continuous improvement of monitoring systems and user feedback mechanisms, platforms can impose immediate penalties on content entrepreneurs who violate GenAI policies. Therefore, sanction celerity can serve as a deterrent.

Based on the studies by Johnston, Warkentin, and Siponen [36] and Burns et al. [40], this study defines perceived sanction celerity as content entrepreneurs’ subjective assessment of the swiftness with which punishment would be imposed following the detection of their GenAI abuse. Immediate punishment allows content entrepreneurs to fully recognize the platform’s ability to regulate both creators and content. This continuously improving regulatory mechanism compels content entrepreneurs to comply with GenAI policies during the content creation process. From a behavioral economics perspective [17,48], immediate punishment raises the immediate cost of violations, intensifying time-related opportunity costs and diminishing returns for content entrepreneurs. Consequently, this time-sensitive cost calculus motivates content entrepreneurs to comply with GenAI policies. Guided by these insights, the following hypothesis is formulated.

H3: 

Perceived sanction celerity for GenAI abuse positively affects content entrepreneurs’ intention to comply with GenAI policies.

The exploratory interview findings indicate that, alongside the deterrence triad, social norm also plays a significant role in shaping policy compliance intention, which aligns with Herath and Rao [17]. Following Chen et al. [11], this study defines perceived social norm in GenAI-driven content entrepreneurship as content entrepreneurs’ belief that the majority of their peers comply with GenAI policies, and such compliance is recognized and encouraged by platforms, peers, and audiences.

Within the creator economy, which thrives on social connectivity and collaborative communities [20], content entrepreneurs actively participate in closely connected peer networks. Through systematically observing and analyzing peers’ GenAI-driven content strategies and their outcomes, they develop clear understandings of shared behavioral standards. Building on the social learning theory [51], when content entrepreneurs perceive strong social norm favoring responsible GenAI use—observing peers adhering to policies while receiving positive reinforcement (e.g., benefits, trust) or witnessing negative consequences for violations (e.g., sanctions, credibility erosion)—they internalize these norms. This internalization, driven by desires for social approval, positive self-image, and avoidance of community disapproval, fosters normative commitment and professional duty [17,52]. Consequently, high perceived social norm has the potential to directly strengthen policy compliance intention. Therefore, the following hypothesis is derived.

H4: 

Perceived social norm positively affects content entrepreneurs’ intention to comply with GenAI policies.

This study also proposes that perceived social norm can interact with all sanction factors to influence content entrepreneurs’ willingness to comply with GenAI policies, drawing on Chen et al. [11]. This theoretical integration explicitly addresses the rapidly evolving GenAI landscape, where formal regulatory frameworks often lag behind technological advancements, creating a governance gap. Social norm—a community-embedded adaptive force—functions as a dynamic complementary mechanism that bridges this gap by fostering voluntary GenAI policy compliance [46].

Following social learning theory [51], under a high level of perceived social norm, content entrepreneurs internalize GenAI policy compliance as a normative imperative and expression of group identity. Driven by this conformity acquired through observation, content entrepreneurs sustain policy compliance regardless of low sanction certainty, severity, or celerity, as they prioritize social approval and professional reputation. Conversely, in an environment with weak social norms, content entrepreneurs observe peers engaging in non-compliant behaviors with minimal social repercussions, leading them to perceive compliance as non-essential for social acceptance. Consequently, the intrinsic motivation to adhere to policies diminishes, making external sanction mechanisms more effective in enforcing compliance. Fundamentally, perceived social norm attenuates the deterrent effects of sanctions by fostering self-sustaining compliance through observational learning and social reinforcement [11]. These arguments align with Paternoster and Simpson [53], whose rational choice perspective demonstrates that cost–benefit calculations become secondary when behavior is reinforced by internalized norms and social rewards, whereas individuals rely more heavily on sanction threats when such normative reinforcement is absent. Building on these analyses, the following hypotheses frame perceived social norm as a moderating variable.

H5: 

Perceived social norm weakens the effect of perceived sanction certainty on content entrepreneurs’ intention to comply with GenAI policies.

H6: 

Perceived social norm weakens the effect of perceived sanction severity on content entrepreneurs’ intention to comply with GenAI policies.

H7: 

Perceived social norm weakens the effect of perceived sanction celerity on content entrepreneurs’ intention to comply with GenAI policies.

Meanwhile, based on the qualitative study findings, peer communication emerges as a critical antecedent to perceptions of the deterrence triad and social norm. In social media-based consumer behavior research, peer communication refers to consumers’ interactions about products/services via computer-mediated social networks [54]. Aligned with this definition, this study defines peer communication as content entrepreneurs’ interactions and information exchanges regarding how to better utilize GenAI through computer-mediated social networks. Theoretical and empirical evidence suggests that peer communication influences behavior through two distinct pathways: informational influence, which alters cognitive assessments, and normative influence, which shapes perceived social expectations [55].

As an informational mechanism, peer communication alters individuals’ cognition by providing accessible knowledge and enabling vicarious learning [51,56]. This mechanism is empirically supported by findings in consumer behavior, where peer communication enhances responsiveness to product advertising [54] and similarly shape product judgments by providing sufficient information [55]. This logic finds further support in the domain of IS security policy compliance, where corporate education programs enhance employees’ perceptions of sanction certainty and severity through knowledge dissemination [12]. Extending this established evidence to GenAI-driven content entrepreneurship, peer communication can elevate content entrepreneurs’ perceptions of sanction certainty, severity, and celerity by disseminating concrete enforcement knowledge (e.g., penalty cases, detection mechanisms) and enabling vicarious learning. Consequently, this study proposes the following three hypotheses.

H8: 

Peer communication positively affects perceived sanction certainty.

H9: 

Peer communication positively affects perceived sanction severity.

H10: 

Peer communication positively affects perceived sanction celerity.

As a normative mechanism, peer communication plays a pivotal role in shaping norms for GenAI usage. According to Bandura’s [51] social learning theory, individuals acquire social norm through information exchange and direct observation of others’ behaviors. Interpersonal information exchanges enable individuals to understand peers’ attitudes and behaviors toward specific objectives, allowing them to infer group norms [57]. Within social media environments, peer communication serves as a critical medium for transmitting behavioral information and social norm to individuals [58]. This normative transmission mechanism operates with particularly high salience among content creator communities, where peers exchange rich GenAI-related knowledge encompassing not only technical skills but also ethical requirements for GenAI utilization. Through sustained informational influence stemming from peer communication, content entrepreneurs gradually develop stable social norms regarding GenAI usage. This study therefore proposes the following hypothesis.

H11: 

Peer communication positively affects perceived social norm.

4.2. Research Method

This confirmatory study employs an online survey to empirically test the research hypotheses. A survey instrument is designed to obtain data on six research variables. Measurement items are adapted from existing literature with minor modifications to align with the context of this study. All constructs are assessed using multi-item Likert scales ranging from 1 (strongly disagree) to 7 (strongly agree). The complete measurement items and corresponding sources are detailed in

Table 3. Measurement items.

Constructs	Items	Sources
Peer Communication	1. I discuss GenAI usage with my peers.	[56]
	2. I ask my peers for advice about GenAI usage.
	3. I obtain information about GenAI usage from my peers.
	4. I talk with my peers about using GenAI for content creation.
Perceived Sanction Certainty	1. I am likely to incur sanctions if I violate GenAI policies.	[15,40]
	2. Sanctions will follow if GenAI policies are violated.
	3. If caught committing a GenAI policy violation, the probability of sanction would be high.
	4. It is likely that I would be punished if I were caught violating GenAI policies.
Perceived Sanction Severity	1. It is likely that the punishment would be severe if I violate GenAI policies.	[15,40]
	2. Sanctions for violations of GenAI policies would be severe.
	3. If I were caught violating GenAI policies, the sanctions would be very severe.
	4. If I violate GenAI policies, the sanctions would put me in serious trouble.
Perceived Sanction Celerity	1. The punishment from GenAI policy violation would be swift.	[36,40]
	2. I would be punished quickly for GenAI policy violation.
	3. Sanctions for GenAI policy violation would be delivered quickly.
	4. Punishment to GenAI policy violations would be instantaneous.
Perceived Social Norm	1. I believe that other peers comply with the GenAI policies.	[11,17]
	2. It is likely that the majority of other peers comply with the GenAI policies.
	3. I am convinced that other peers comply with the GenAI policies.
Policy Compliance Intention	1. I intend to comply with the requirements of GenAI polices in the future.	[15,16]
	2. I intend to perform my responsibilities prescribed in the GenAI policies.
	3. I am likely to follow the GenAI policies.
	4. I intend to comply with the GenAI policies.

.

4.3. Data Collection

Since the unit of analysis in this study is content entrepreneurs who used GenAI for content creation, obtaining an adequate sample size of such entrepreneurs poses significant challenges. To address this issue, we adopted a convenience sampling strategy and engaged Wenjuanxing, a leading professional data collection platform in China, to conduct the questionnaire distribution and data collection. The data collection procedure consists of three sequential steps: First, the survey is designed on Wenjuanxing, incorporating predefined scales adapted to the research context. Then, the questionnaire is commissioned to Wenjuanxing for distribution and collection, with data collection required to be completed within one month. Finally, rigorous screening measures are implemented to ensure data quality and verify respondents’ identities as genuine content entrepreneurs.

In the respondent identity verification phase, a four-step confirmation process is implemented. First, at the beginning of the questionnaire, respondents are explicitly informed that participation is restricted to content entrepreneurs who used GenAI for content creation. Second, respondents are required to report the names of their most frequently used GenAI tools. Third, detailed instructions are provided for respondents to demonstrate their status as content entrepreneurs through the submission of personal content account screenshots. Finally, these screenshots are required to clearly display the content entrepreneur’s published content alongside engagement metrics—such as likes, subscriptions, audience size (followers), and/or revenue sharing—serving as valid evidence for identity verification. A reward of ¥40 is offered to respondents who complete the questionnaire appropriately. Within a month, 186 questionnaires were collected; after eliminating 19 unqualified questionnaires, a total of 167 valid questionnaires were obtained. This sample, though constrained by recruitment challenges, meets and exceeds the 10-times rule threshold for PLS-SEM [59], ensuring statistical validity.

Table 4. Demographics of respondents (n = 167).

Category	Item	Frequency	Percentage
Gender	Male	112	67.1
	Female	55	32.9
Age	18–19	4	2.4
	20–29	83	49.7
	30–39	62	37.1
	>39	18	10.8
Education	High school or lower	29	17.4
	Bachelor’s or college degree	135	80.8
	Graduate degree	3	1.8
Industry Sector	Service	50	29.9
	Manufacturing	37	22.2
	Agriculture	3	1.8
	Student	16	9.6
	Others	61	36.5
Duration of Content Creation	≤12 Months	6	3.6
	13–24 Months	62	37.1
	25–36 Months	71	42.5
	>36 Months	28	16.8
Frequently Used GenAI	Doubao	87	52.1
	deepseek	24	14.4
	ERNIE	14	8.4
	ChatGPT	8	4.8
	Others	34	20.3
Total	-	167	100

summarizes respondent demographics.

4.4. Data Analysis and Results

PLS-SEM demonstrates particular suitability for managing non-normally distributed data [60] and effectively supports analyses with sample sizes below 500 observations [61]. Given these advantages, this study utilizes Smart PLS 4.0 to evaluate measurement reliability, confirm convergent validity, verify discriminant validity, check common method bias (CMB), and statistically test the proposed hypotheses.

First, confirmatory factor analysis is performed to evaluate the measurement model’s reliability and validity. As shown in

Table 5. Results of reliability and convergent validity tests.

Construct	Indicator	Standardized Loading	CR	AVE	Cronbach’s α
Peer Communication	PC1	0.843	0.858	0.696	0.853
	PC2	0.881
	PC3	0.854
	PC4	0.753
Perceived Sanction Certainty	SCer1	0.837	0.884	0.738	0.882
	SCer2	0.870
	SCer3	0.881
	SCer4	0.849
Perceived Sanction Severity	SSev1	0.887	0.912	0.786	0.909
	SSev2	0.894
	SSev3	0.893
	SSev4	0.871
Perceived Sanction Celerity	SCel1	0.881	0.876	0.724	0.873
	SCel2	0.883
	SCel3	0.813
	SCel4	0.825
Perceived Social Norm	PSN1	0.896	0.858	0.778	0.857
	PSN2	0.878
	PSN3	0.872
Policy Compliance Intention	PCI1	0.895	0.912	0.789	0.911
	PCI2	0.872
	PCI3	0.903
	PCI4	0.884

, consistent with the criteria established by Fornell and Larcker [62], the Cronbach’s α value and the composite reliability (CR) value for all constructs surpass the threshold value of 0.700, indicating acceptable levels of internal consistency and scale reliability. Regarding convergent validity, in accordance with Fornell and Larcker’s [62] criteria, all measurement items demonstrated standardized factor loadings substantially exceeding 0.700, while all constructs yielded average variance extracted (AVE) estimates surpassing the 0.500 threshold, thereby confirming convergent validity.

Second, discriminant validity is assessed using Fornell and Larcker’s [62] criterion, which involves comparing the square root of the AVE for each construct against the inter-construct correlations.

Table 6. Construct correlations and discriminant validity.

Construct	Mean	S.D.	1	2	3	4	5	6
1. Perceived Sanction Celerity	5.400	0.987	0.851
2. Perceived Sanction Certainty	5.523	1.047	0.608	0.859
3. Policy Compliance Intention	5.647	0.961	0.583	0.597	0.888
4. Peer Communication	5.801	1.034	0.622	0.564	0.611	0.834
5. Perceived Sanction Severity	5.421	1.095	0.133	0.093	0.269	0.140	0.886
6. Perceived Social Norm	5.597	0.925	0.581	0.597	0.572	0.533	0.148	0.882

The diagonal numbers in bold are the square roots of the AVE.

presents the correlation matrix, with the square roots of AVE (bolded diagonal entries) and inter-construct correlations, facilitating a comparative assessment of discriminant validity. Every bolded diagonal value (square root of AVE) exceeds all corresponding row and column correlations in the matrix, thus confirming discriminant validity. Additionally, the heterotrait–monotrait ratio (HTMT) proposed by Hair et al. [59] is employed to further verify discriminant validity. As shown in

Table 7. Heterotrait–Monotrait (HTMT) ratio of correlations.

Construct	1	2	3	4	5	6
1. Perceived Sanction Celerity	-
2. Perceived Sanction Certainty	0.691	-
3. Policy Compliance Intention	0.653	0.665	-
4. Peer Communication	0.720	0.648	0.694	-
5. Perceived Sanction Severity	0.150	0.104	0.295	0.159	-
6. Perceived Social Norm	0.666	0.683	0.647	0.620	0.169	-

, the maximum HTMT value is 0.720, which falls below the threshold of 0.900, providing definitive evidence for the presence of discriminant validity.

Third, given the potential for CMB in self-reported single-source data, this study conducts Harman’s one-factor test to assess CMB concerns following MacKenzie and Podsakoff [63]. Analysis indicates that there are six latent factors with eigenvalues exceeding 1.0, and the first factor accounts for 41.99% of total variance—below the 50% critical threshold. This confirms that CMB does not substantially compromise the study’s validity.

Finally, to examine the proposed hypotheses, this study employs Smart PLS 4.0 to conduct a path analysis. The findings from hypothesis testing are illustrated in Figure 2. With respect to the relationships between deterrence triad and policy compliance intention, perceived sanction certainty, perceived sanction severity, and perceived sanction celerity each exhibit significant positive impacts on GenAI policy compliance intention (β = 0.231, p < 0.01; β = 0.151, p < 0.01; β = 0.197, p < 0.01, respectively), thereby confirming support for H1, H2, and H3.

Regarding the relationship between perceived social norm and policy compliance intention, the result demonstrates that perceived social norm exerts a positive effect on policy compliance intention (β = 0.175, p < 0.01), thus supporting H4. Meanwhile, this study employs a two-stage PLS approach [59] to examine interaction effects (i.e., perceived sanction certainty × perceived social norm, perceived sanction severity × perceived social norm, and perceived sanction celerity × perceived social norm) on policy compliance intention. Bootstrapping analysis using 5000 resamples reveals that perceived social norm significantly attenuates the relationship between perceived sanction certainty and policy compliance intention (β = −0.202, p < 0.01). Conversely, the perceived social norm does not significantly moderate the relationship between perceived sanction severity and policy compliance intention (β = 0.034, p > 0.05) or between perceived sanction celerity and policy compliance intention (β = 0.012, p > 0.05). These results confirm that H5 is supported while H6 and H7 are rejected.

Furthermore, results confirm that peer communication significantly strengthens perceived sanction certainty (β = 0.564, p < 0.001), perceived sanction celerity (β = 0.622, p < 0.001), and perceived social norm (β = 0.533, p < 0.001), but not perceived sanction severity (β = 0.140, p > 0.05). Thus, H8, H10, and H11 are supported whereas H9 is rejected.

In addition, this study investigates the mediating roles of perceived sanction certainty, perceived sanction severity, perceived sanction celerity, as well as perceived social norm in the proposed model. Following Hair et al.’s [59] analytical guidelines, the findings reveal that, with the exception of H9 (peer communication → perceived sanction severity), all direct effects (H1–H4, H8, H10, H11) are statistically significant. Meanwhile, the direct effect (peer communication → policy compliance intention, controlling for mediators) is significant (β = 0.367, p < 0.001); the indirect pathways—peer communication → perceived sanction certainty → policy compliance intention (β = 0.130, p < 0.01), peer communication → perceived sanction celerity → policy compliance intention (β = 0.122, p < 0.05), and peer communication → perceived social norm → policy compliance intention (β = 0.093, p < 0.05)—are also statistically significant and directionally consistent with theoretical predictions. These results confirm that perceived sanction certainty, perceived sanction celerity, and perceived social norm act as partial mediators in the model.

5. Discussion and Implications

5.1. Key Findings

This study employs a mixed-methods approach to examine content entrepreneurs’ compliance with GenAI policies. First, the results indicate that peer communication strengthens compliance intention by enhancing perceived sanction certainty and celerity, while showing no significant effect on sanction severity. This pattern contrasts with findings on institutional deterrence mechanisms. For instance, D’Arcy, Hovav, and Galletta [12] demonstrate that security policies, SETA programs, and computer monitoring significantly affect both sanction severity and certainty. Raddatz, Marett, and Trinkle [16] confirm these relationships while also showing that computer monitoring does not improve sanction celerity. The divergence in pathways indicates that sanction severity perceptions appear rooted in direct experiences with institutional enforcement, as reflected by Participant #6: “The platform requires me to prove that the content is my original work; otherwise, my account will be throttled. That’s really troublesome! So, I have no choice but to play by its rules.” In contrast, peer communication directly shapes perceptions of sanction certainty and celerity through shared experiences and informal discussions. These findings establish peer communication as a distinct governance mechanism that activates a different combination of deterrent factors, differing from the mechanisms found in studies of structured interventions.

Second, this study establishes perceived social norm as a mediator between peer communication and GenAI policy compliance intention. While Chen et al. [11] demonstrate that formal deterrence measures reinforce social norms through fear-driven inferences in organizational settings, this research reveals that informal peer communication directly shapes normative perceptions regarding GenAI use in content entrepreneurship environments. This contrast positions peer communication as a distinct, bottom-up pathway to social norm formation.

Third, this study finds that perceived social norm interacts with perceived sanction certainty, but not with severity or celerity, to influence compliance intention. This result aligns with Chen et al. [11], which also finds no interaction between perceived descriptive norm and sanction severity. The lack of significant interactions for severity and celerity stems from their objectively defined and system-inherent nature. Sanction severity is concretely established by predefined penalties in platform policies (e.g., a 30-day account suspension for posting non-compliant GenAI content). Likewise, sanction celerity is principally determined by the platform’s technical enforcement capacity (e.g., automatically issuing a penalty within 48 h of detecting policy-violating GenAI content). Because the levels of sanction severity and celerity are largely fixed and externally defined by the platform, content entrepreneurs’ perceptions of them are formed by directly assessing these explicit institutional rules and technical specifications. Thus, the cost–benefit calculation arising from these relatively objective and stable factors remains separate from the socially constructed influence of perceived social norm, providing a clear explanation for the non-significant interaction effects.

In contrast, perceived sanction certainty is shaped by both official initiatives (e.g., platform pop-ups alerting content entrepreneurs to recent GenAI violation penalties) and social cues (e.g., content entrepreneurs sharing screenshots of account warnings). Since both official channels and social interactions transmit information about enforcement likelihood, this shared susceptibility to social influence creates a common ground with perceived social norm, enabling their interaction. Specifically, when entrepreneurs perceive strong peer compliance, their motivation shifts from fear of detection to alignment with group expectations, thus weakening the role of sanction certainty. Conversely, when perceived social norm is weak, their compliance intention becomes more dependent on their perception of sanction certainty, thereby amplifying its effect. This dynamic is well illustrated by quotes from Participant #4: “There’s a large volume of GenAI-generated deceptive product recommendation content on the platform that attracts significant traffic. I also want to do this!… But now, things have changed. The platform has initiated a stringent crackdown on GenAI-generated fraudulent content.”

5.2. Theoretical Implications

This study may make several theoretical contributions to the research community. First, while GenAI-enabled content entrepreneurship has expanded rapidly, it has concurrently revealed substantial unresolved challenges and critical knowledge gaps that necessitate urgent scholarly attention [23,24]. Specifically, on the supply side, the increasing abuse of GenAI by content entrepreneurs may trigger a creator dilemma [22] and value co-destruction [27]. Existing literature has primarily addressed these issues through conceptual research [8,26] and case studies [29,64]. To bridge this gap, this study advances a methodological contribution by employing a mixed-methods approach to empirically uncover mitigation strategies for content entrepreneurs’ GenAI abuse. The design uses exploratory qualitative interviews to identify key factors and confirmatory quantitative surveys to validate their mechanisms, thereby providing a methodological and theoretical foundation for mitigating GenAI abuse in content entrepreneurship.

Second, prior research has predominantly centered on deterrence theory in domains such as crime prevention [48,65] and IS security [16,36,40]. This study is among the first to apply the general deterrence theory to examine content entrepreneurs’ compliance with GenAI policies, with a particular emphasis on the previously understudied role of perceived sanction celerity. The validation of all three deterrence factors confirms and refines the theory’s applicability to GenAI governance, demonstrating its effectiveness in regulating content entrepreneurship. This study responds to the appeal by Laine, Minkkinen, and Mäntymäki [8] for establishing a theoretical grounding in GenAI governance. By systematically examining how deterrence theory influences content entrepreneurs’ compliance with GenAI policies, this research offers critical theoretical support for developing effective GenAI regulatory systems. The refinement of this theoretical foundation serves as an essential prerequisite for promoting the responsible and efficient adoption of GenAI across diverse domains [26,29,66].

Third, grounded in the community-driven nature of content entrepreneurship [20], this study advances understanding of how perceived social norm interacts with deterrence mechanisms. It addresses limitations in prior research that treated perceived social norm primarily as an independent variable [17] or examined its interaction effect only with sanction severity [11]. The results demonstrate that perceived social norm selectively moderates the influence of perceived sanction certainty—but not severity or celerity—on compliance intention. This precise delineation of significant interaction effects clarifies the boundary conditions governing the interplay between perceived social norm and perceived sanctions in digital content ecosystems, thereby providing a critical theoretical foundation for designing collaborative governance models that integrate platform-led deterrence with community-based self-regulation.

Finally, this study establishes peer communication as a novel governance avenue for GenAI by demonstrating its capacity to address the limitations of top-down regulation through two distinct pathways—sharpening deterrence perceptions while cultivating social norms. By extending peer communication from consumer behavior research [54,55,56] to the domain of content entrepreneurs’ GenAI policy compliance, this research provides new theoretical grounding for multi-stakeholder governance [66]. This community-embedded, socially interactive approach complements existing regulatory frameworks by offering a more adaptive and decentralized alternative to traditional governance paradigms.

5.3. Practical Implications

This study offers four key practical implications. First, platforms should strategically leverage peer communication to enhance content entrepreneurs’ awareness of sanction certainty and celerity, as these factors significantly improve compliance intention. An effective approach is to establish peer-led forums that facilitate discussions on GenAI governance and share authentic case studies. For example, Bilibili has created content entrepreneur communities where members actively share their experiences with GenAI usage. These discussions typically cover proper usage techniques while also presenting real penalty cases involving GenAI abuse. Such community-based initiatives make the likelihood of detection and speed of penalties more concrete and credible than top-down communication alone. This peer-driven approach complements formal platform efforts to enhance monitoring transparency [29], effectively promoting responsible GenAI adoption among content entrepreneurs.

Second, platforms must proactively strengthen the deterrence of sanction severity, which directly enhances compliance intention even though it is less susceptible to peer influence. Consistent with D’Arcy, Hovav, and Galletta [12], this perception of sanction severity stems from content entrepreneurs’ direct assessment of penalty harshness rather than indirect peer influence. To capitalize on this insight, platforms should develop clear, tiered penalty frameworks under the guidance of national regulations and in partnership with governing bodies. Such a system would issue warnings for minor violations, temporarily remove content for repeated offenses, and suspend accounts for severe abuses. For example, content platforms such as Xiaohongshu and Douyin, in compliance with relevant GenAI governance laws and regulations issued by the Chinese government, have successively launched multiple rounds of special governance campaigns. They have removed tens of thousands of non-compliant GenAI-generated content items and banned a large number of accounts that seriously violated GenAI policies, effectively curbing content entrepreneurs’ tendency to abuse GenAI for content creation and dissemination.

Third, fostering ongoing peer communication to shape positive social norms represents a crucial strategy for strengthening GenAI policy compliance. Platforms should establish concrete mechanisms that make compliant behavior visible and incentivized through structured recognition systems. For instance, Douyin’s Creator Honor Points system publicly acknowledges content creators who consistently follow platform policies, while Xiaohongshu’s Creator Credit System maintains transparent scoring based on compliance records. These reputation mechanisms should be integrated with substantive benefits—including prioritized content distribution and enhanced recommendation visibility. When content creators actively share and recognize these rewarded compliance behaviors within their networks, individual cases become observable social proof that establishes clear benchmarks for normative conduct. This approach reduces reliance on top-down regulation and fosters self-regulatory cultures of compliance.

Finally, to optimize GenAI governance, platforms should actively identify the social norms prevalent among content creators regarding GenAI usage, then strategically integrate peer communication with regulatory measures. For mature communities with well-established self-regulatory practices, platforms should prioritize guidance over intensive monitoring, allocating resources primarily to support community self-governance through structured programs that facilitate peer-led compliance mentoring. For newly established creator groups still developing normative standards for GenAI usage, platforms should implement balanced strategies combining regulatory oversight with normative cultivation. For instance, following Chen et al. [11], platforms like WeChat, Douyin, and Xiaohongshu can regularly publish content governance announcements to strengthen the deterrent effect of regulations. Concurrently, platforms should establish peer education mechanisms through which experienced creators share practical insights to foster awareness of responsible GenAI usage.

5.4. Limitations and Future Research Directions

While this study offers valuable insights into the drivers of content entrepreneurs’ compliance with GenAI policies, it is important to acknowledge its limitations. First, it relies solely on deterrence theory as the theoretical basis to explore mechanisms for promoting content entrepreneurs’ GenAI policy compliance intention. However, prior research on employees’ compliance with IS policies has often integrated deterrence theory with other theories and perspectives, such as neutralization theory [38], institutional theory [12], and self-regulation perspectives [13], to construct more comprehensive theoretical models. These studies provide valuable theoretical frameworks that can be adapted to research on content entrepreneurs’ compliance with GenAI policies. Future research could further employ mixed-methods approaches to more appropriately reveal the methods and pathways for increasing content entrepreneurs’ compliance with GenAI policies by leveraging theoretical integration.

Second, as this study focuses on Chinese content entrepreneurs, the generalizability of its findings to other cultural contexts remains an open question requiring further validation. Research such as Men and Muralidharan [56] has demonstrated that cultural factors can influence the dynamics and outcomes of peer communication. Given that content entrepreneurship is currently a prevailing global trend [3], developing a GenAI policy compliance framework with universal applicability is highly imperative. It is therefore recommended that future research conduct cross-cultural comparative studies, particularly between collectivist and individualist societies, to enhance generalizability and elucidate culture’s moderating role.

Third, a potential limitation concerns the exploratory qualitative phase of this mixed-methods study. Consistent with established guidelines for mixed-methods research [18], the interviews are designed to identify key concepts and/or theories. Although interviews with seven content entrepreneurs adequately served this purpose and reached theoretical saturation, the relatively small sample size may still limit the breadth and diversity of perspectives captured. To enhance the robustness of qualitative insights, future research could employ a larger and more diverse sample or adopt a longitudinal interview design. Such approaches would help identify potential novel factors that shape content entrepreneurs’ compliance with GenAI policies.

Fourth, during the confirmatory research phase, this study relied on surveys for data collection, which may raise concerns about CMB [63]. Although statistical tests indicate that CMB does not pose a significant threat in this study, future research is encouraged to employ alternative data collection methods (e.g., experimental designs, data mining) to more comprehensively assess content entrepreneurs’ behaviors and psychological characteristics. Additionally, it would be valuable to examine the influence of demographic variables on compliance intention, which could further enhance the robustness and generalizability of the findings.

Finally, this study did not limit data collection to a specific content platform or GenAI tool, primarily due to the challenges in obtaining large-scale samples of content entrepreneurs. As different platforms and GenAI tools implement distinct GenAI policies, this approach in the current study may lead to variations in content entrepreneurs’ understanding of sanction measures and policy compliance intention. Future research is recommended to collect data from a specific platform and, more significantly, to conduct comparative studies across different platforms. Such comparative work would yield more robust evidence for applying deterrence theory to GenAI governance in content entrepreneurship.